Lodestar.Metrics 0.3.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

One 2×2 matrix per label or per sample — sklearn.metrics.multilabel_confusion_matrix.

public static ConfusionMatrix[] Compute(ReadOnlySpan<int> yTrue, ReadOnlySpan<int> yPred, ReadOnlySpan<int> labels = default, ReadOnlySpan<double> sampleWeight = default)
public static ConfusionMatrix[] Compute(ReadOnlySpan<bool> yTrue, ReadOnlySpan<bool> yPred, int labelCount, bool samplewise = false, ReadOnlySpan<double> sampleWeight = default)

Parameters — the first overload takes yTrue and yPred as one label per sample and reports one matrix per class, with labels fixing which classes and in what order; omit it for the sorted union of both inputs. The second takes them as a row-major label matrix of labelCount values per row, and samplewise decides whether to count one matrix per label or one per row. sampleWeight is one weight per sample — per row, not per label.

Returns — a fresh ConfusionMatrix[]: one per class in label order, one per label in column order, or one per sample in row order.

Exceptions — ArgumentException when the inputs disagree in length, are empty, when the matrix is not a whole number of rows of labelCount, or when the weights do not match the sample count.

Example — three labels over two samples.

using Lodestar.Metrics;

bool[] truth = [true, false, true, false, true, true];
bool[] predicted = [true, false, false, true, true, true];

ConfusionMatrix[] perLabel = MultilabelConfusionMatrix.Compute(truth, predicted, labelCount: 3);
int matrices = perLabel.Length;  // => 3

Counting the same input by sample instead returns one matrix per row:

using Lodestar.Metrics;

bool[] truth = [true, false, true, false, true, true];
bool[] predicted = [true, false, false, true, true, true];

ConfusionMatrix[] perSample = MultilabelConfusionMatrix.Compute(truth, predicted, 3, samplewise: true);
int matrices = perSample.Length;  // => 2

Remarks — each entry is an ordinary ConfusionMatrix over labels 0 and 1, so Recall.Score, Precision.Score and the rest read it directly. Under samplewise a row's weight applies to each of that row's labels, because the matrix counts labels there rather than samples.

Applies to — net10.0, netstandard2.0.

See also — ConfusionMatrix.Compute, Precision.PerClass, the Python equivalence table.